Re: st: Do I use survey? with suest? Follow-up question

[Steven Samuels <samplerx@earthlink.net>]

On Dec 18, 2006, at 1:16 AM, Pura Rayco-Solon wrote:
_____________________________________________
Two provinces were chosen, one as control area and the
other as experimental area.
From each province, 3 cities were chosen.
From each city, 3 villages were chosen.
From each village, a random sample of women were
chosen - the number of women chosen from each village
was proportional to the population size of the
village.

The weighting does not make a difference on either the
point estimate nor the standard error but the
clustering does (and the difference becomes
non-significant).
_________________________________________
Pura:

You will probably need stata's multi-level commands, like xtmixed and  
gllamm, not the survey commands.

On rereading your description, I am not sure that you took a  
probability ("random") sample of each province, or even of the  
province's cities. A random sample would have selected cities at  
random from a list, and villages at random from the list. To compute  
a sampling weight, you would need the probability of selection at  
each stage.  A minor issue is that, even with a probability sample,  
the weight at the village level would be
(village size/sample size), not village size.

Even with  a full-fledged random-sampling design at all stages, you   
you have run into the issue that Stata's survey analyses are "design- 
based," and  the effective "sample size"  is essentially the number  
of PSU's in this case 6!   That is why your results become non- 
significant when city is the PSU.

To advise you better, we would need a detailed description of the  
study and intervention.

Steve




*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/